zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-09-01 12:49:50 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Switch to fast RMS/LN Norm (#603)

Browse Source 
* use nn.RMSNorm, use sdpa, cleanup

* bump mlx versions

* minor update

* use fast layer norm

* version bump

* update requirement for whisper

* update requirement for gguf
This commit is contained in:
Awni Hannun
2024-03-23 07:13:51 -07:00
committed by GitHub
parent fbed720d6f
commit b8a348c1b8
44 changed files with 144 additions and 1155 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
llms/llama/llama.py
View File

@@ -28,20 +28,6 @@ class ModelArgs:
rope_traditional: bool = True
class RMSNorm(nn.Module):
def __init__(self, dims: int, eps: float = 1e-5):
super().__init__()
self.weight = mx.ones((dims,))
self.eps = eps
def _norm(self, x):
return x * mx.rsqrt(x.square().mean(-1, keepdims=True) + self.eps)
def __call__(self, x):
output = self._norm(x.astype(mx.float32)).astype(x.dtype)
return self.weight * output
class Attention(nn.Module):
def __init__(self, args: ModelArgs):
super().__init__()
@@ -120,8 +106,8 @@ class TransformerBlock(nn.Module):
self.dim = args.dim
self.attention = Attention(args)
self.feed_forward = FeedForward(args=args)
self.attention_norm = RMSNorm(args.dim, eps=args.norm_eps)
self.ffn_norm = RMSNorm(args.dim, eps=args.norm_eps)
self.attention_norm = nn.RMSNorm(args.dim, eps=args.norm_eps)
self.ffn_norm = nn.RMSNorm(args.dim, eps=args.norm_eps)
self.args = args
def __call__(
@@ -144,7 +130,7 @@ class Llama(nn.Module):
self.vocab_size = args.vocab_size
self.tok_embeddings = nn.Embedding(args.vocab_size, args.dim)
self.layers = [TransformerBlock(args=args) for _ in range(args.n_layers)]
self.norm = RMSNorm(args.dim, eps=args.norm_eps)
self.norm = nn.RMSNorm(args.dim, eps=args.norm_eps)
self.output = nn.Linear(args.dim, args.vocab_size, bias=False)
def __call__(self, x):